[REGRESSION][BISECTED] Freeze when nouveau loads GSP firmware during boot on Blackwell

[REGRESSION][BISECTED] Freeze when nouveau loads GSP firmware during boot on Blackwell

[Matthew Schwartz]

Hello,

On the 6.19-rc kernel series with linux-firmware 20251125, I have been unable to boot into the kernel while using nouveau on my RTX 5090 rig. 

It appears to freeze right after "nouveau 0000:01:00.0: gsp: RM version: 570.144" appears in my dmesg, and it locks up at this point. It works fine if I blacklist nouveau or use nomodeset.

After bisecting, the first bad commit is da67179e5538 ("drm/nouveau/gsp: Allocate fwsec-sb at boot"). As it's a Blackwell card, I cannot test with GSP disabled.

Still trying to figure out the best way to provide complete logs of the freeze, as it's too early to go into my journal and sysrq keys don't work when it freezes... 

Thanks,
Matt

RE: [REGRESSION][BISECTED] Freeze when nouveau loads GSP firmware during boot on Blackwell

[Christopher Snowhill]

[-- Attachment #1: Type: text/plain, Size: 1495 bytes --]



On Tue 30 Dec 2025 10:11:34 PM , Matthew Schwartz wrote:
> Hello,
> 
> On the 6.19-rc kernel series with linux-firmware 20251125, I have been unable to boot into the kernel while using nouveau on my RTX 5090 rig. 
> 
> It appears to freeze right after "nouveau 0000:01:00.0: gsp: RM version: 570.144" appears in my dmesg, and it locks up at this point. It works fine if I blacklist nouveau or use nomodeset.
> 
> After bisecting, the first bad commit is da67179e5538 ("drm/nouveau/gsp: Allocate fwsec-sb at boot"). As it's a Blackwell card, I cannot test with GSP disabled.
> 
> Still trying to figure out the best way to provide complete logs of the freeze, as it's too early to go into my journal and sysrq keys don't work when it freezes... 
> 
> Thanks,
> Matt

I encountered the same bug possibly, except in my case, it resulted in
an OOPS, and the total failure of video output. However, I was able to
log into LUKS and pull some kernel logs:

https://lore.kernel.org/all/176698808133.6372.2408917375327107249@copycat/

The commonality here seems to be crashing dereferencing offset 0xCC in a
structure from bit_entry(struct nvkm_bios *bios, u8 id, struct bit_entry *bit)

Nearest I can guess without doing further proper examination is that
it's dereferencing a NULL nvkm_bios pointer right at the top of the
function. Maybe something isn't initializing the BIOS member before this
ctor is called on Blackwell? Race conditions are really fun to deal
with.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

RE: [REGRESSION][BISECTED] Freeze when nouveau loads GSP firmware during boot on Blackwell

[Christopher Snowhill]

[-- Attachment #1: Type: text/plain, Size: 1495 bytes --]



On Tue 30 Dec 2025 10:11:34 PM , Matthew Schwartz wrote:
> Hello,
> 
> On the 6.19-rc kernel series with linux-firmware 20251125, I have been unable to boot into the kernel while using nouveau on my RTX 5090 rig. 
> 
> It appears to freeze right after "nouveau 0000:01:00.0: gsp: RM version: 570.144" appears in my dmesg, and it locks up at this point. It works fine if I blacklist nouveau or use nomodeset.
> 
> After bisecting, the first bad commit is da67179e5538 ("drm/nouveau/gsp: Allocate fwsec-sb at boot"). As it's a Blackwell card, I cannot test with GSP disabled.
> 
> Still trying to figure out the best way to provide complete logs of the freeze, as it's too early to go into my journal and sysrq keys don't work when it freezes... 
> 
> Thanks,
> Matt

I encountered the same bug possibly, except in my case, it resulted in
an OOPS, and the total failure of video output. However, I was able to
log into LUKS and pull some kernel logs:

https://lore.kernel.org/all/176698808133.6372.2408917375327107249@copycat/

The commonality here seems to be crashing dereferencing offset 0xCC in a
structure from bit_entry(struct nvkm_bios *bios, u8 id, struct bit_entry *bit)

Nearest I can guess without doing further proper examination is that
it's dereferencing a NULL nvkm_bios pointer right at the top of the
function. Maybe something isn't initializing the BIOS member before this
ctor is called on Blackwell? Race conditions are really fun to deal
with.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]